Cubature formulas, geometrical designs, reproducing kernels, and Markov operators
نویسندگان
چکیده
Cubature formulas and geometrical designs are described in terms of reproducing kernels for Hilbert spaces of functions on the one hand, and Markov operators associated to orthogonal group representations on the other hand. In this way, several known results for spheres in Euclidean spaces, involving cubature formulas for polynomial functions and spherical designs, are shown to generalize to large classes of finite measure spaces (Ω, σ) and appropriate spaces of functions inside L2(Ω, σ). The last section points out how spherical designs are related to a class of reflection groups which are (in general dense) subgroups of orthogonal groups. Let (Ω, σ) be a finite measure space and let F be a vector space of integrable real-valued functions on Ω. It is a natural question to ask when and how integrals ∫ Ω φ(ω)dσ(ω) can be computed, or approximated, by sums ∑ x∈X W (x)φ(x), where X is a subset of Ω and W : X −→ R+ a weight function, for all φ ∈ F . When Ω is an interval of the real line, this is a basic problem of numerical integration with a glorious list of contributors: Newton (1671), Cotes (1711), Simpson (1743), Gauss (1814), Chebyshev (1874), Christoffel (1877), and Bernstein (1937), to quote but a few; see [Gauts–97] for some historical notes, [Gauts–76], and [DavRa–84]. The theory is inseparable of that of orthogonal polynomials [Szegö–39]. When Ω is a space of larger dimension, the problems involved are often of geometrical and combinatorial interest. One important case is that of spheres in Euclidean spaces, with rotation-invariant probability measure, as in [DeGoS–77] and [GoeSe–79]. Work related to integrations domains with dim(Ω) > 1 goes back to [Maxwe–77] and [Appel–90]; see also the result of Voronoi (1908) recalled in Item 1.15 below. Examples which have been considered include various domains in Euclidean spaces (hypercubes, simplices, ..., Item 1.20), surfaces (e.g. tori), and Euclidean spaces with Gaussian measures (Item 3.11). There are also interesting cases where Ω itself is a finite set [Delsa–78]. Section 1 collects the relevant definitions for the general case (Ω, σ). It reviews several known examples on intervals and spheres. Our main point is to show that there are two notions which are convenient for the study of cubature formulas, even if they are rarely explicitely introduced in papers of this subject. First, we introduce in Section 2 the formalism of reproducing kernel Hilbert spaces ([Arons–50], [Krein–63], [BeHaG–03]), which is appropriate for generalizing to other spaces 1991 Mathematics Subject Classification. Primary 05B30, 65D32. Secondary 46E22.
منابع مشابه
Lagrange polynomials, reproducing kernels and cubature in two dimensions
We obtain by elementary methods necessary and sufficient conditions for a k-dimensional cubature formula to hold for all polynomials of degree up to 2m− 1 when the nodes of the formula have Lagrange polynomials of degree at most m. The main condition is that the Lagrange polynomial at each node is a scalar multiple of the reproducing kernel of degree m− 1 evaluated at the node plus an orthogona...
متن کاملNumerical performance of optimized Frolov lattices in tensor product reproducing kernel Sobolev spaces
In this paper, we deal with several aspects of the universal Frolov cubature method, that is known to achieve optimal asymptotic convergence rates in a broad range of function spaces. Even though every admissible lattice has this favorable asymptotic behavior, there are significant differences concerning the precise numerical behavior of the worst-case error. To this end, we propose new generat...
متن کاملMinimal cubature formulae of trigonometric degree
In this paper we construct minimal cubature formulae of trigonometric degree: we obtain explicit formulae for low dimensions of arbitrary degree and for low degrees in all dimensions. A useful tool is a closed form expression for the reproducing kernels in two dimensions.
متن کاملNumerical Cubature from Archimedes' Hat-box Theorem
Archimedes’ hat-box theorem states that uniform measure on a sphere projects to uniform measure on an interval. This fact can be used to derive Simpson’s rule. We present various constructions of, and lower bounds for, numerical cubature formulas using moment maps as a generalization of Archimedes’ theorem. We realize some well-known cubature formulas on simplices as projections of spherical de...
متن کاملLower bounds for projective designs, cubature formulas and related isometric embeddings
Yudin’s lower bound [21] for the spherical designs is generalized to the cubature formulas on the projective spaces over a field K ⊂ {R, C, H} and thus to isometric embeddings l 2;K → l p;K with p ∈ 2N. For large p and in some other situations this is essentially better than those known before. AMS Classification: 46B04, 05B30
متن کامل